October 15, 2026

Hand an AI agent one whole job this week. Keep the approvals.

OpenAI says small businesses are moving from asking AI questions to handing agents whole tasks. Here is how to try one job with ChatGPT Work, and the three checks to make first.

Consultiply is an OpenAI Select Partner. This is our owner's read of what changed and how to try it safely.

On October 14, 2026, OpenAI published new research showing that businesses are moving from asking AI questions to letting agents carry out tasks "from planning through completion." The fastest growth it reports is among small businesses. This post explains what OpenAI found, what its agent tool does, and how to try handing it one job this week without giving up control.

What OpenAI published

OpenAI's Economic Research team looked at how European companies on its business plans use ChatGPT. Two numbers stand out, both from OpenAI's own customer data:

More than half of the work is agentic. In OpenAI's EU sample of ChatGPT Enterprise customers, Codex and ChatGPT Work, its two agent tools, "now account for more than half of output tokens," the units AI models process and generate.

Small businesses are catching up fast. Since May 2026, weekly active users of those tools at EU small and medium businesses, which OpenAI defines as 200 or fewer users, "grew more than eightfold, with growth across every member state." OpenAI's chart puts it at 8.1x.

OpenAI line chart titled 'Agentic weekly active users in EU SMBs grew 8.1x since May', rising from a 1.0x baseline on May 7 to 8.1x on Oct 5, with the steepest climb from mid-July. — Chart: OpenAI, from its Oct 14, 2026 Economic Research post on X. OpenAI's own customer data.

The pattern is not only European. In its Sep 30 small business report, OpenAI said agentic output from products like ChatGPT Work and Codex made up two-thirds of all small business output tokens in August, up from one-third in April. It also found that small-business users lean on agents far more than larger firms for finance and accounting, supply chain, legal and compliance, and marketing, the jobs a small team often has no specialist for.

These are OpenAI's numbers about its own paying customers. They show how quickly use is growing, not whether it paid off for those businesses.

What ChatGPT Work is, in plain words

ChatGPT Work, launched on July 9, 2026, is an agent inside ChatGPT. Instead of answering one question, it takes a goal, gathers information from the apps and files you connect, breaks the job into steps and returns finished material: sheets, slides, docs or a simple web page. OpenAI says you can follow its progress, answer its questions, change direction "and approve important actions" as it works. Scheduled Tasks let it repeat a job on a schedule or when something changes.

Who gets it, and where it works

01
Plans.
OpenAI rolled ChatGPT Work out on web and mobile to Pro, Enterprise and Edu first, then Plus and Business over the following days. In the ChatGPT desktop app, Chat, Work and Codex are available on every plan, including Free. Features can reach accounts at different times, so check yours.
02
Usage.
OpenAI says ChatGPT Work "is designed for longer, more involved work than a typical chat request, so usage works differently." Bigger jobs use more of your plan's included usage.
03
Connections.
It works with what you connect through plugins, such as email, calendars, Google Drive or SharePoint, and CRMs. It can only use what you give it.

Four jobs to try

Note: these are our examples, not OpenAI screens. Each uses something OpenAI documents ChatGPT Work doing. Test with sample or low-stakes data first.

01
Competitor price check.
"Look up current prices for [your top five products or services] at these competitors: [names and websites]. Put them in a sheet next to my prices and flag where I'm more than 10% above or below." What you get: a comparison sheet you can act on. Tip: OpenAI describes a New Mexico restaurant owner doing exactly this before raising menu prices.
02
Month-end budget review.
"Using the attached budget and this month's export from my accounting tool, list every line that's over or under budget by more than 10% and suggest a one-line reason for each." What you get: a short variance list. Tip: OpenAI suggests starting with a task you already know well, and month-end numbers are one you can check yourself.
03
Customer feedback themes.
"Every Monday, read last week's customer reviews and support emails, group them into themes and give me the top three with example quotes." What you get: a weekly summary through Scheduled Tasks. Tip: start with one source, such as reviews, before connecting your inbox.
04
Sales meeting prep.
"I'm meeting [customer] on Thursday. Pull our past emails, open quotes and notes, and give me a one-page brief with their open questions." What you get: a brief before the call. Tip: keep it read-only; let it draft, not send.

Three checks before you rely on it

01
Check the numbers before a customer sees them.
Agents work fast and can still get a figure or a source wrong. Spot-check every number against the original.
02
Decide what your team may use it for.
Write down which jobs are fine to hand over and which actions always need a person's approval: sending messages, paying, changing records or prices.
03
Be careful with customer and employee details.
Connect only the apps the job needs, and keep payroll, health or personal records out until you know where the data goes.

Our take

The shift OpenAI describes is real in its data: small businesses are handing agents whole jobs, not just questions. Start with one familiar job, keep the approvals with you, and widen what it can do only after it has earned it.